Scoring systems for upper gastrointestinal bleeding: Which one scores better?

Early warning system enables accurate mortality risk prediction for acute gastrointestinal bleeding admitted to intensive care unit

Acute gastrointestinal (GI) bleeding are potentially life-threatening conditions. Early risk stratification is important for triaging patients to the appropriate level of medical care and intervention. Patients admitted to intensive care unit (ICU) has a high mortality, but risk tool is scarce for these patients. This study aimed to develop and validate a risk score to improve the prognostication of death at the time of patient admission to ICU. We developed and internally validated a nomogram for mortality in patients with acute GI bleeding from the eICU Collaborative Research Database (eICU-CRD), and externally validated it in patients from the Medical Information Mart for Intensive Care III database (MIMIC-III) and Wuhan Tongji Hospital. The performance of the model was assessed by examining discrimination (C-index), calibration (calibration curves) and usefulness (decision curves). 4750 patients were included in the development cohort, with 1184 patients in the internal validation cohort, 1406 patients in the MIMIC-III validation cohort, and 342 patients in the Tongji validation cohort. The nomogram, which incorporated ten variables, showed good calibration and discrimination in the training and validation cohorts, yielded C-index ranged from 0.832 (95%CI 0.811-0.853) to 0.926 (95CI% 0.905-0.947). The nomogram-defined high-risk group had a higher mortality than the low-risk group (44.8% vs. 3.5%, P < 0.001; 41.4% vs 3.1%, P < 0.001;53.6% vs 7.5%, P < 0.001; 38.2% vs 4.2%, P < 0.001). The model performed better than the conventional Glasgow-Blatchford score, AIMS65 and the newer Oakland and Sengupta scores for mortality prediction in both the derivation and validation cohorts concerning discrimination and usefulness. Our nomogram is a reliable prognostic tool that might be useful to identify high-risk acute GI bleeding patients admitted to ICU.

The Accuracy of Pre-Endoscopic Scores for Mortality Prediction in Patients with Upper GI Bleeding and No Endoscopy Performed

(1) Background: The assessment of mortality and rebleeding rate in upper gastrointestinal bleeding (UGIB) is essential, and several prognostic scores have been proposed. Some patients with UGIB did not undergo endoscopy, either because they refused the procedure, suffered from alcohol withdrawal symptoms or altered general status, or because the bleeding was severe enough to cause death before the endoscopy. The mortality risk in the subgroup of patients without endoscopy is poorly evaluated in the literature. (2) Methods: The purpose of the study was to identify the most useful scores for the assessment of in-hospital mortality in patients with UGIB with no endoscopy performed and no known etiology. A total of 198 patients with UGIB and no endoscopy performed were admitted between January 2017 and December 2021 and the accuracy of 12 prognostic scores and the Charlson comorbidity index for in-hospital mortality prediction were analyzed, as well as Child-Pugh Turcotte (CPT) and Meld scores in patients with cirrhosis. (3) Results: The mortality rate was 37.9%, higher than in variceal (21.9%, p < 0.0001) and non-variceal bleeding (7.4%, p < 0.0001). The most accurate scores by AUC were the International Bleeding score (INBS, 0.844), Glasgow Blatchford (0.783), MAP score (0.78), Iino (0.766), AIM65 and modified N-score (0.745 each), modified Glasgow-Blatchford (0.73), H3B2 and N-score (0.701); Rockall, Baylor, and T-score had an AUC below 0.7. MELD score was superior to CPT in patients with cirrhosis (AUC 0.811 versus 0.670). (4) Conclusions: The mortality rate in UGIB with no endoscopy was higher than in both variceal and non-variceal bleeding and was higher in the pandemic period but with no statistical significance (45.3% versus 32.14%, p = 0.0586), mainly because of positive cases. Only one case of rebleeding was noted; the hospitalization period was significantly shorter. The most accurate score was International Bleeding Score; the MELD score had a higher but moderate accuracy compared with CPT in patients with cirrhosis.

Explainable Machine Learning Model for Predicting GI Bleed Mortality in the Intensive Care Unit

INTRODUCTION

Acute gastrointestinal (GI) bleed is a common reason for hospitalization with 2%-10% risk of mortality. In this study, we developed a machine learning (ML) model to calculate the risk of mortality in intensive care unit patients admitted for GI bleed and compared it with APACHE IVa risk score. We used explainable ML methods to provide insight into the model's prediction and outcome.

METHODS

We analyzed the patient data in the Electronic Intensive Care Unit Collaborative Research Database and extracted data for 5,691 patients (mean age = 67.4 years; 61% men) admitted with GI bleed. The data were used in training a ML model to identify patients who died in the intensive care unit. We compared the predictive performance of the ML model with the APACHE IVa risk score. Performance was measured by area under receiver operating characteristic curve (AUC) analysis. This study also used explainable ML methods to provide insights into the model's outcome or prediction using the SHAP (SHapley Additive exPlanations) method.

RESULTS

The ML model performed better than the APACHE IVa risk score in correctly classifying the low-risk patients. The ML model had a specificity of 27% (95% confidence interval [CI]: 25-36) at a sensitivity of 100% compared with the APACHE IVa score, which had a specificity of 4% (95% CI: 3-31) at a sensitivity of 100%. The model identified patients who died with an AUC of 0.85 (95% CI: 0.80-0.90) in the internal validation set, whereas the APACHE IVa clinical scoring systems identified patients who died with AUC values of 0.80 (95% CI: 0.73-0.86) with P value <0.001.

DISCUSSION

We developed a ML model that predicts the mortality in patients with GI bleed with a greater accuracy than the current scoring system. By making the ML model explainable, clinicians would be able to better understand the reasoning behind the outcome.